[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: hdb stops checkpointing....

> --On Monday, November 13, 2006 12:32 PM +0100 johan.jonemo@hep.lu.se
> wrote:
>> I try to modify a hdb periodically (often). After a while the underlying
>> database seems to be swamped down.
>> If I look in the directory log files are piling up. I run db_archive
>> periodically to clean up the log files and I also run db_deadlock in
>> parallel.
>> I do not run db_checkpoint as this as far as I understand is done by the
>> backend.
>> Strangely the problem seems to start after about 2 h if I start from a
>> clean database. When I try to lower the frequency of the updates it
>> still
>> seems to happen after 2 h.
>> I use openLDAP v2.3.24
>> with Sleepycat Software: Berkeley DB 4.4.20
> If you set the right BDB options in DB_CONFIG, I can't see any reason to
> run any of the db_* commands.

Then there must be some serious bug in back-hdb because I haven't seen it
delete a single log-file. Wether the db is responsive or not no
(berkley-db-)log files seems to be erased.

My starting point was not to run any "db_"-commands. But as the directory
seems to be acting up I have to try every possibility.

The db-commands seem to delay the breakdown slightly, but I can't tell
because I'm also testing other alternatives. The latest breakdown took 3
hours. The Breakdown times might of course be random. I haven't had time
to statistically establish that the breakdown times tend to any one value.

here is my DB_CONFIG:
set_cachesize 0 16777216 8
set_lg_regionmax 262144
set_lg_bsize 2097152
set_lg_max 16777216

It should be noted that the directory is at no time expected to take on
very large sizes. Otherwise some memory sizes might have been set slightly

Johan "the one who'll lose his job if he doesn't get this to work" Jönemo